Does mind perception explain the uncanny valley effect? A meta-regression analysis and (de)humanization experiment.

Abstract

Gray and Wegner (2012) proposed that when robots look human, their appearance prompts attributions of experience, including sensations and feelings, which is uncanny. This theory, confusingly termed mind perception, differs from perceptual theories of the uncanny valley in that the robots' eeriness is not stimulus-driven. To explore this seminal theory, we conducted a meta-regression analysis of 10 experiments and a (de)humanization experiment. In the first part, experiments were identified in the literature that manipulated artificial entity's experience using descriptions. However, experiments with no observable stimuli yielded larger effects for experience and eeriness than those with robots and virtual reality characters. This finding undermines a theory that purports to explain how a robot's human likeness causes eeriness. Further, a second issue concerns Gray and Wegner's protocol based on a vignette design. Reading about an entity with experience activates thoughts that may not be activated when encountering it, and these thoughts may increase its eeriness. Therefore, the paper's second part focuses on an experiment we conducted with a novel humanization–dehumanization protocol. Participants' attitudes on robots' similarity to humans were gradually shifted to manipulate robots' perceived humanness, experience, and agency. However, the manipulation's effect on eeriness and coldness was mostly nonsignificant or counter to prediction. Differences in the robots' physical appearance had a much larger effect on their eeriness and coldness. In fact, as a mediator, experience mitigated the stimulus's overall effect of increasing eeriness. These results favor perceptual theories, rather than mind perception, in explaining the uncanny valley.